Strategy and Application of Data-Driven Testing of an Ocean Turbine Drivetrain

Abstract

An ocean turbine generator will be subject to variations in water current velocity at one end and resistive load at the other. Measurements acquired during bench testing of the turbine's drivetrain can be used to predict the effect of these variables on the turbine during deployment. a This paper outlines an ongoing series of tests involving vibration data captured at various velocities and loads. This data is analyzed to detect the machine's state using two approaches. In the first approach, machine learning techniques are used to discern between states given adjacent values for velocity or load. The purpose of that approach is to assess a learner's ability to draw the fine distinctions needed for subsequent fault identification. The second approach computes a power spectrum for each velocity and load combination over enough trials as to construct an operating envelope. The purpose of that approach is to identify as abnormal, incoming data having parts of its spectrum that fall outside that envelope. A case study applies each approach to baseline data and data for a deterioration scenario. The rationale behind each approach, their assumptions, and their limitations are also discussed.